Scheduling activity sales volume prediction method and device

Abstract

The invention provides a schedule activity sales volume prediction method and device. The method comprises the following steps: selecting historical transaction data of high-sale commodities to construct basic data; the construction features comprise activity reference sales features; adopting a machine learning tree model algorithm, and taking the daily average sales volume of each commodity participating in the activity in the current schedule as a target variable; grouping according to the activity types, and training in each group by adopting a corresponding group training set to obtain a group prediction model; classifying commodities participating in the to-be-predicted activity into common commodities and new commodities; and according to the type of the to-be-predicted activity, selecting the corresponding grouping prediction model to obtain a daily average sales prediction value of the common commodity. By accurately predicting the daily average sales volume of the commodities in the activity period, the daily average sales volume can be multiplied by the number of days of the activity to obtain the total sales volume in the whole activity period, the schedule sales volume prediction error of the commodities participating in the activity is reduced, then the working process of a purchasing department is optimized, and the operation efficiency and customer satisfaction of a store are improved.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a method and device for predicting sales volume of a schedule activity. Background Art

[0002] Currently, large-scale commodity sales platforms have diversified commodity sales channels, aiming to widely meet the needs of various consumers. However, during each sales period, the sales performance of stores in various cities presents a complex situation, and the types of activities in which commodities participate are also diverse, which leads to backlogs or out-of-stock situations for some commodities. This phenomenon may not only lead to waste of commodity resources, thereby increasing the management costs of stores, but also make it difficult for consumers to obtain satisfactory commodities during regular shopping hours.

[0003] It provides a sales forecasting method for period activities, which can predict the sales of merchandise participating in future store periods, thereby providing the purchasing department with a scientific and reasonable basis for merchandise procurement, and helping stores to arrange merchandise according to their respective characteristics, which plays a significant role in optimizing the operations of each store. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and device for predicting sales volume during a schedule activity, which are used to predict the sales volume of merchandise participating in activities at each store during future schedules.

[0005] In order to solve the above problems, the present invention provides a method for predicting sales of activities during a schedule, including: obtaining historical transaction data of a target commodity to construct basic data, wherein the target commodity is a commodity with a sales volume higher than a preset value among all commodities; constructing features, wherein the features include activity reference sales volume features, and the activity reference sales volume features include: the average daily sales volume and the maximum sales volume of each commodity in various types of activities in the past preset time period, and the average daily sales volume and the maximum sales volume of each commodity in the most recent activity; separating a training process from a prediction process, executing a training process once on the first day of the current schedule, and after the training process is completed, executing a prediction process once on each day of the current schedule to predict the average daily sales volume of each commodity in the activity to be predicted; the training process includes: using a machine learning tree model algorithm to classify each commodity participating in the activity in the current schedule The average daily sales volume is used as the target variable; the basic data is grouped into a group training set according to the type of activity, and the corresponding group training set is used in each group to train the tree model to obtain a group prediction model; model evaluation and attribution analysis are automatically performed; the evaluation results are uploaded to the intelligent data analysis platform and the prediction ability of the group prediction model is displayed on the interface; the prediction process includes: classifying the commodities participating in the activity to be predicted into ordinary commodities and new products; for the ordinary commodities, according to the type of the activity to be predicted, the corresponding group prediction model is selected to predict the average daily sales volume forecast value, and for the new products, the average daily sales volume of the commodities under the commodity subcategory to which the new products belong is used as the average daily sales volume forecast value; the average daily sales volume forecast values ​​of all ordinary commodities and the average daily sales volume forecast values ​​of all new products are integrated as the prediction result.

[0006] In some embodiments, the construction of the activity reference sales feature in the feature construction step adopts the following steps in sequence: if a specific activity type does not appear in a specific commodity in the basic data, the relevant data of the commodity subcategory to which the specific commodity belongs is used to construct the activity reference sales feature of the specific commodity in the specific activity; if there is no relevant data in the commodity subcategory, the relevant data of the commodity subcategory to which the specific commodity belongs is used to construct the activity reference sales feature of the specific commodity in the specific activity; if there is no relevant data in the commodity subcategory, the relevant data of other subcategories with the specific activity is used to construct the activity reference sales feature of the specific commodity in the specific activity; if there is no relevant data in the other subcategories, the relevant data of other subcategories with the specific activity is used to construct the activity reference sales feature of the specific commodity in the specific activity; if there is no relevant data in the other subcategories, the historical sales average of the specific commodity in other types of activities is used to construct the activity reference sales feature of the specific commodity in the specific activity.

[0007] In some embodiments, the features further include: holiday features, activity-related features, display location features, product attribute features, and date features.

[0008] In some embodiments, the activity-related characteristics include type-level characteristics of the promotion activity, price difference characteristics, commodity purchase limit characteristics, and discount intensity characteristics.

[0009] In some embodiments, the holiday characteristics include characteristics related to traditional festivals and characteristics related to shopping festivals, and the activity reference sales characteristics also include reference sales characteristics of festive activities related to traditional festivals and shopping festivals.

[0010] In some embodiments, the training process further includes removing outliers in the group training set.

[0011] In some embodiments, the activities include marketing activities and promotional activities, and the training process further includes: sampling and processing the data of the marketing activities in the basic data.

[0012] In some embodiments, the prediction process further includes: exporting the prediction results to a database. Exporting the prediction results to a database.

[0013] In some embodiments, the forecast result is multiplied by the number of days of the activity to be forecasted to obtain the sales forecast value of the activity to be forecasted.

[0014] In order to solve the above problems, the present invention also provides a sales forecasting device for a schedule activity, including: a data processing module, the data processing module acquires historical transaction data of a target commodity to construct basic data, the target commodity being a commodity with a sales volume higher than a preset value among all commodities; a feature construction module, the feature construction module is used for feature construction, the feature includes the feature includes activity reference sales volume feature, the activity reference sales volume feature includes: the average daily sales volume and maximum sales volume of each commodity in various types of activities in the past preset time period, and the average daily sales volume and maximum sales volume of each commodity in the most recent activity; a training module, the training module is used to train and generate a group prediction model, the training module executes a training process: using a machine learning tree model, taking the average daily sales volume of each commodity participating in the activity in the current schedule as the target variable, grouping the basic data according to the type of activity to form a group training set, and using the corresponding group training set in each group to train the tree model respectively The model is used to obtain the group prediction model, automatically perform model evaluation and attribution analysis, and upload the evaluation results to the intelligent data analysis platform to intuitively understand the model prediction ability on the interface; the prediction module, the prediction module is used to predict the average daily sales of each commodity in the activity to be predicted, and the prediction module executes the prediction process: the commodities participating in the activity to be predicted are classified into ordinary commodities and new products. For the ordinary commodities, according to the type of the activity to be predicted, the corresponding group prediction model is selected to predict the average daily sales forecast value; for the new products, the average daily sales of the commodities under the commodity subcategory to which the new products belong is used as the average daily sales forecast value, and the average daily sales forecast values ​​of all ordinary commodities and the average daily sales forecast values ​​of all new products are integrated as the prediction result; the scheduling module, the scheduling module calls the training module on the first day of the current period to execute the training process once, and calls the prediction module on each day of the current period to execute the prediction process once.

[0015] The above technical scheme constructs basic data by acquiring historical transaction data of target commodities, wherein the target commodities are commodities whose sales volume is higher than a preset value among all commodities; constructs features including activity reference sales volume features; executes a training process on the first day of the current period: adopts a machine learning tree model algorithm, and takes the average daily sales volume of each commodity participating in the activity in the current period as the target variable; groups the basic data into a group training set according to the type of activity, and uses the corresponding group training set to train each group to obtain a group prediction model; executes a prediction process on each day of the current period: classifies the commodities participating in the activity to be predicted into ordinary commodities and new products; for the ordinary commodities, selects the corresponding group prediction model according to the type of the activity to be predicted to predict the average daily sales volume forecast value, and for the new products, takes the average daily sales volume of the commodities under the commodity subcategory to which the new products belong as the average daily sales volume forecast value; integrates the average daily sales volume forecast values ​​of all ordinary commodities and the average daily sales volume forecast values ​​of all new products as the prediction result. By accurately predicting the average daily sales of goods during an event, the average daily sales volume can be multiplied by the number of days of the event to obtain the total sales volume during the entire event, thereby reducing the sales forecast error of the products participating in the event, thereby optimizing the workflow of the purchasing department and improving the store's operational efficiency and customer satisfaction.

[0016] It should be understood that the above general description and the detailed description below are exemplary and explanatory only and cannot limit the present invention. The techniques, methods and devices known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the authorization specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments of the present invention are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 This is a flow chart of a method for predicting sales volume of a schedule activity provided by an embodiment of the present invention;

[0019] Figure 2 is a flow chart of a training process provided by an embodiment of the present invention;

[0020] Figure 3 is a flowchart of a prediction process provided by an embodiment of the present invention;

[0021] Figure 4 It is a schematic diagram of a sales forecasting device for a schedule activity provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] The activities described in the present invention refer to marketing activities or sales promotion activities in commodity sales.

[0024] Please refer to Figure 1 to Figure 3 , Figure 1 is a flowchart of a method for predicting sales volume of a schedule activity provided by an embodiment of the present invention. Figure 2 is a flowchart of a training process provided by an embodiment of the present invention, Figure 3 It is a flowchart of a prediction process provided by an embodiment of the present invention.

[0025] like Figure 1 As shown, the sales prediction method for a period activity includes: step S11, obtaining historical transaction data of a target commodity to construct basic data, wherein the target commodity is a commodity with sales higher than a preset value among all commodities; step S12, feature construction, wherein the feature includes activity reference sales features, and the holiday features include features related to traditional festivals and features related to shopping festivals. The activity reference sales features include the average sales and maximum sales of various activity types in the past two years, as well as the average sales and maximum sales of the most recent activity type; step S13, training process; step S14, prediction process.

[0026] Referring to step S11, the historical transaction data of the target commodity is obtained to construct basic data, and the target commodity is a commodity with a sales volume higher than a preset value among all commodities. Specifically, this step classifies the commodities according to their sales volume based on the historical transaction data of all commodities, and classifies the commodities with a sales volume higher than a preset value as target commodities, and classifies the commodities with a sales volume lower than or equal to the preset value as low-selling commodities. In order to optimize the training efficiency and effectiveness of the model, the present invention does not use the sales volume prediction method of the schedule activity to predict the low-selling commodities, nor does it use the historical transaction data of the low-selling commodities to construct the basic data. Instead, the historical transaction data of the target commodity is used to construct the basic data. The preset value is a certain sales volume value of the commodity.

[0027] In this embodiment, the commodities whose sales volume does not exceed 1 in each period of the most recent year are classified as the low-selling commodities, and the remaining commodities are classified as the target commodities, and the preset value is 1.

[0028] In this step, the historical transaction data is omni-channel transaction data, including online transaction data and offline transaction data.

[0029] The basic data includes schedule data and schedule sales data. The current schedule data includes the number of days the schedule lasts and special schedules. The current schedule data also includes the effective time span of each activity in the schedule, which is used to indicate that the schedule and the activity validity period do not completely overlap. For example, the duration of the schedule is 14 days, but some activities are only effective within 7 days before or after the schedule.

[0030] Referring to step S12, feature construction, the feature includes an activity reference sales feature, and the activity reference sales feature includes: the average daily sales and maximum sales of each commodity in various types of activities in the past preset time period, and the average daily sales and maximum sales of each commodity in the most recent activity. The preset time period refers to a specific time period. In this embodiment, the preset time period is 2 years. Accordingly, the activity reference sales feature includes: the average daily sales and maximum sales of each commodity in various types of activities in the past 2 years, and the average daily sales and maximum sales of each commodity in the most recent activity.

[0031] The activity reference sales feature is constructed based on the historical activity data of the product. However, in the historical data of the product, some activity types sometimes do not appear, which makes it impossible to directly construct the activity reference sales feature.

[0032] In this step, if a specific activity type does not appear for a specific commodity in the basic data, the relevant data of the commodity sub-category to which the specific commodity belongs is used to construct the activity reference sales characteristics of the specific commodity in the specific activity; if there is no relevant data for the commodity sub-category, the relevant data of the commodity sub-category to which the specific commodity belongs is used to construct the activity reference sales characteristics of the specific commodity in the specific activity; if there is no relevant data for the commodity sub-category, the relevant data of other sub-districts with the specific activity is used to construct the activity reference sales characteristics of the specific commodity in the specific activity; if there is no relevant data for the other sub-districts, the historical sales average of the specific commodity in other types of activities is used to construct the activity reference sales characteristics of the specific commodity in the specific activity. The specific commodity refers to any commodity in the basic data, and the specific activity refers to any activity for which sales forecasting is required. Through this series of alternative measures, it is ensured that a comprehensive and effective activity reference sales characteristics can be constructed even in the absence of direct data.

[0033] In this step, feature engineering focuses on constructing features that are closely related to sales. In this embodiment, the features also include holiday features, which include features involving traditional festivals and features involving shopping festivals. The traditional festival feature refers to whether the activity involves traditional festivals such as "Spring Festival, Mid-Autumn Festival, Dragon Boat Festival, Labor Day, Tomb-Sweeping Day, Women's Day, National Day". The shopping festival feature refers to whether the activity involves informal shopping-related holidays such as "Double 11", "Double 12", and "618".

[0034] In this embodiment, activities with festivals and activities without festivals are distinguished, and activities with festivals are added with reference sales features of festivals when constructing features, thereby providing a more accurate and reliable reference factor for the model.

[0035] In this embodiment, the features further include: activity-related features, display location features, product attribute features, and date features.

[0036] The activity-related features are used to reflect the characteristics of commodity activities and related details, ensuring that the model can accurately grasp the comprehensive impact of activities on sales. The activity-related features include the type and level features of the promotion activities, the price difference features, the commodity purchase limit features, and the discount strength features.

[0037] The display location feature is used to reflect the potential impact of the product placement of the store during the event on sales. The product attribute feature is used to reflect the inherent attributes of the product itself. The date feature is used to reflect the sales change pattern of periodic products, and the date feature includes the number of days of vacation during the event.

[0038] In order to optimize the utilization of computing resources and separate the training and prediction processes, the training process is separated from the prediction process. The training process is executed once on the first day of the current period, and after the training process is completed, the prediction process is executed once on each day of the current period to predict the average daily sales of each commodity in the activity to be predicted.

[0039] In this embodiment, the training process is executed starting at 0:00 on the first day of the current period, and the prediction process is executed at 6:00 in the morning of each day of the current period to predict the average daily sales of each commodity in the to-be-predicted activity.

[0040] Refer to step S13 and Figure 2The training process includes: step S131, using a machine learning tree model algorithm, taking the average daily sales of each commodity participating in the activity during the current period as the target variable; step S132, grouping the basic data according to the type of activity to form a group training set, and using the corresponding group training set in each group to train the tree model to obtain a group prediction model; step S133, automatically performing model evaluation and attribution analysis; step S134, uploading the evaluation results to the intelligent data analysis platform and displaying the prediction ability of the group prediction model on the interface.

[0041] Referring to step S131, a machine learning tree model algorithm is used to take the average daily sales of each commodity participating in the activity during the current period as the target variable. The average daily sales is the final output of the generated prediction model. The total sales of each commodity during the activity can be obtained by multiplying the average daily sales of each commodity by the number of days of the activity. In this embodiment, the number of days is determined in combination with the region, store, and start and end time of the activity, so that the sales of the active commodities during the period can be accurately calculated. This series of steps ensures the accuracy of the data and the depth of business analysis.

[0042] Referring to step S132, the basic data is grouped according to the type of activity to form a group training set, and the corresponding group training set is used in each group to train the tree model to obtain a group prediction model. In order to ensure that the model can accurately capture the sales volume change trend of the product under different activity types and avoid using a single model to process all data, in this embodiment, the basic data is grouped according to the type of marketing activity and the type of promotion activity to form a group training set, and the model is independently trained in each group to obtain a group prediction model. A total of 28 groups of group prediction models are trained, so that more accurate predictions can be made for products of different activity types.

[0043] In this step, outliers in the group training set are removed. Outliers refer to values ​​that are too large or too small and are obviously beyond the range.

[0044] Among them, the activity types include marketing activity types and promotional activity types. The marketing activity types further include store modifications and online advertising. The promotional activity types further include discount activities, full-reduction activities, gift activities, member discount activities, limited-time special activities and consumer rebate activities.

[0045] In view of the fact that there are many types of marketing products, huge data volume, and limited computing resources, this step also samples the marketing data in the basic data. The sampling strategy is based on the principle of product diversification and is carried out according to the store product dimension.

[0046] Referring to step S133, model evaluation and attribution analysis are automatically performed. This step is used to ensure real-time feedback on the model prediction effect and is automatically performed on the first day of each schedule. Among them, the attribution analysis results can provide feedback on the performance of the group prediction model, which can then be used to accurately adjust the model training strategy.

[0047] Referring to step S134, the evaluation result is uploaded to the intelligent data analysis platform and the prediction capability of the group prediction model is displayed on the interface. In this embodiment, the intelligent data analysis platform is Quick_BI.

[0048] Refer to step S14 and Figure 3 The prediction process includes: step S141, classifying the commodities participating in the activity to be predicted into common commodities and new products; step S142, for the common commodities, according to the type of the activity to be predicted, selecting the corresponding group prediction model to predict the average daily sales forecast value, and for the new products, taking the average daily sales of the commodities under the commodity subcategory to which the new products belong as the average daily sales forecast value; step S143, integrating the average daily sales forecast values ​​of all common commodities and the average daily sales forecast values ​​of all new products as the prediction result.

[0049] Referring to step S141, the commodities participating in the activity to be predicted are classified into common commodities and new commodities, wherein the new commodities refer to commodities sold on the shopping platform for the first time, and the common commodities refer to commodities that have been sold on the shopping platform.

[0050] Referring to step S142, for the common commodity, according to the type of the activity to be predicted, the corresponding group prediction model is selected to predict the average daily sales forecast value, and for the new product, the average daily sales of the commodities under the commodity subcategory to which the new product belongs is used as the average daily sales forecast value. The group prediction model is trained based on the historical transaction data of existing commodities. When the new product is predicted by the model, there is a large uncertainty in accuracy. The new product is predicted by a specific rule method, which can ensure the robustness of the new product prediction result.

[0051] Referring to step S143, the average daily sales forecast values ​​of all common commodities and the average daily sales forecast values ​​of all new products are integrated as the forecast results. In addition, the forecast results are exported to the database. In this embodiment, the forecast results are also exported to Oracle and ES databases for reference and use by downstream parties.

[0052] The prediction result is the average daily sales of the products participating in the activity during the current period. The prediction result is multiplied by the number of days of the activity to be predicted to obtain the sales forecast value of the activity to be predicted, which can eliminate the potential interference and uncertainty caused by the difference in activity duration on the model prediction result. Among them, the prediction result includes the average daily sales of each product, and correspondingly, the sales forecast value includes the predicted total sales of each product.

[0053] Based on the same inventive concept, the present application also provides a sales forecasting device for a schedule event.

[0054] Figure 4 Schematic diagram of a sales forecasting device for a schedule activity provided by an embodiment of the present invention. Figure 4 As shown, the schedule activity sales prediction device includes: a data processing module 21, a feature construction module 22, a training module 23, a prediction module 24, and a scheduling module 25.

[0055] The data processing module 21 acquires historical transaction data of a target commodity to construct basic data, and the target commodity is a commodity whose sales volume is higher than a preset value among all commodities.

[0056] The feature construction module 22 is used for feature construction, and the feature includes an activity reference sales feature, and the activity reference sales feature includes: the average daily sales and maximum sales of each commodity in various types of activities in the past preset time period, and the average daily sales and maximum sales of each commodity in the most recent activity. The preset time period refers to a specific time period. In this embodiment, the preset time period is 2 years.

[0057] The training module 23 is used to train and generate a group prediction model 29. The training module 23 executes a training process: using a machine learning tree model, the average daily sales of each commodity participating in the activity during the current period is used as the target variable, and the basic data is grouped according to the type of activity to form a group training set. In each group, the corresponding group training set is used to train the tree model to obtain a group prediction model 29, automatically perform model evaluation and attribution analysis, and upload the evaluation results to the intelligent data analysis platform for intuitively understanding the model prediction capability on the interface.

[0058] The prediction module 24 is used to predict the average daily sales volume of each commodity in the activity to be predicted. The prediction module 24 executes the prediction process: the commodities participating in the activity to be predicted are classified into ordinary commodities and new products. For the ordinary commodities, according to the type of the activity to be predicted, the corresponding group prediction model 29 is selected to predict the average daily sales volume forecast value. For the new products, the average daily sales volume of the commodities under the commodity subcategory to which the new products belong is used as the average daily sales volume forecast value. The average daily sales volume forecast values ​​of all ordinary commodities and the average daily sales volume forecast values ​​of all new products are integrated as the prediction result.

[0059] The forecast result is multiplied by the number of days of the activity to be forecasted to obtain the sales forecast value of the activity to be forecasted.

[0060] The scheduling module 23 executes the training process once on the first day of the current schedule, and executes the prediction process once on each day of the current schedule after the training process is completed.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion. The various embodiments in this specification are described in a related manner, and the same and similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0062] The above is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. It should be noted that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A sales forecasting method for a schedule activity, characterized in that: include: Obtain historical transaction data of a target product to construct basic data, wherein the target product is a product whose sales volume is higher than a preset value among all products; Feature construction, the feature includes an activity reference sales feature, the activity reference sales feature includes: the average daily sales and maximum sales of each commodity in various types of activities in the past preset time period, and the average daily sales and maximum sales of each commodity in the most recent activity; Separate the training process from the prediction process, execute the training process once on the first day of the current period, and after the training process is completed, execute the prediction process once on each day of the current period to predict the average daily sales of each commodity in the activity to be predicted; The training process includes: Using a machine learning tree model algorithm, the average daily sales of each commodity participating in the activity during the current period is used as the target variable; the basic data is grouped according to the type of activity to form a group training set, and the corresponding group training set is used in each group to train the tree model to obtain a group prediction model; model evaluation and attribution analysis are automatically performed; the evaluation results are uploaded to the intelligent data analysis platform and the prediction ability of the group prediction model is displayed on the interface; The prediction process includes: The commodities participating in the activity to be predicted are classified into common commodities and new products. For the common commodities, according to the type of the activity to be predicted, the corresponding grouping prediction model is selected to predict the average daily sales forecast value. For the new products, the average daily sales of the commodities under the commodity subcategory to which the new products belong is used as the average daily sales forecast value. The average daily sales forecast values ​​of all common commodities and the average daily sales forecast values ​​of all new products are integrated as the forecast result.

2. The method according to claim 1, characterized in that The construction of the activity reference sales feature in the feature construction step adopts the following steps in sequence: If a specific activity type does not appear for a specific commodity in the basic data, then the activity reference sales feature of the specific commodity in the specific activity is constructed using the relevant data of the commodity subcategory to which the specific commodity belongs; If there is no relevant data for the commodity sub-category, then the relevant data of the commodity sub-category to which the specific commodity belongs is used to construct the activity reference sales characteristics of the specific commodity in the specific activity; If there is no relevant data in the category of the commodity, then the relevant data of other communities with the specific activity are used to construct the activity reference sales characteristics of the specific commodity in the specific activity; If there is no relevant data in the other communities, the historical sales average of the specific commodity in other types of activities is used to construct the activity reference sales feature of the specific commodity in the specific activity.

3. The method according to claim 1, characterized in that The features also include: holiday features, activity-related features, display location features, product attribute features, and date features.

4. The method according to claim 3, characterized in that The activity-related characteristics include type and level characteristics of the promotion activity, price difference characteristics, commodity purchase limit characteristics, and discount intensity characteristics.

5. The method according to claim 4, characterized in that The holiday characteristics include characteristics related to traditional festivals and characteristics related to shopping festivals, and the activity reference sales characteristics also include reference sales characteristics of festive activities related to traditional festivals and shopping festivals.

6. The method according to claim 1, characterized in that The training process further includes: removing outliers in the group training set.

7. The method according to claim 1, characterized in that The activities include marketing activities and promotional activities, and the training process further includes: Sampling processing is performed on the data of the marketing activities in the basic data.

8. The method according to claim 1, characterized in that The prediction process further includes: The prediction results are exported to a database.

9. The method according to claim 1, characterized in that: The sales volume forecasting method for a schedule activity further comprises: The forecast result is multiplied by the number of days of the activity to be forecasted to obtain the sales forecast value of the activity to be forecasted.

10. A sales forecasting device for a schedule activity, characterized in that: include: A data processing module, wherein the data processing module acquires historical transaction data of a target commodity to construct basic data, wherein the target commodity is a commodity whose sales volume is higher than a preset value among all commodities; A feature construction module, the feature construction module is used for feature construction, the feature includes an activity reference sales feature, the activity reference sales feature includes: the average daily sales and maximum sales of each commodity in various types of activities in the past preset time period, and the average daily sales and maximum sales of each commodity in the most recent activity; A training module, the training module is used to train and generate a group prediction model, the training module executes a training process: using a machine learning tree model, taking the average daily sales of each commodity participating in the activity in the current period as the target variable, grouping the basic data according to the type of activity to form a group training set, and using the corresponding group training set in each group to train the tree model to obtain a group prediction model, automatically performing model evaluation and attribution analysis, and uploading the evaluation results to the intelligent data analysis platform for intuitively understanding the model prediction ability on the interface; A prediction module, the prediction module is used to predict the average daily sales of each commodity in the activity to be predicted, and the prediction module executes a prediction process: the commodities participating in the activity to be predicted are classified into common commodities and new products, for the common commodities, according to the type of the activity to be predicted, the corresponding group prediction model is selected to predict and obtain the average daily sales forecast value, for the new products, the average daily sales of the commodities under the commodity subcategory to which the new products belong is used as the average daily sales forecast value, and the average daily sales forecast values ​​of all common commodities and the average daily sales forecast values ​​of all new products are integrated as the prediction result; A scheduling module, wherein the scheduling module calls the training module to execute the training process once on the first day of the current schedule, and calls the prediction module to execute the prediction process once on each day of the current schedule.

Images